Drivers of Consumer Resistance to AI

Product/Service InnovationZehnle, M.; Hildebrand, C.; Valenzuela, A. · 2025International Journal of Research in Marketing
Topicsartificial intelligence·ai aversion·meta-analysis·generative ai·algorithms·consumer response·anthropomorphism

Your leadership team is deciding which customer-facing AI services to scale and what assurance each needs. A company-wide assumption about “AI acceptance” hides the variation that matters. Calling the AI a robot draws roughly four times the resistance of calling it an assistant, system, or algorithm, and AI that acts with extensive autonomy draws more resistance than AI with limited autonomy. Transportation, public-safety, and social-welfare uses draw more resistance than operational and lifestyle uses, while the average gap has narrowed over time. Approve the portfolio by use case, with different rollout pace, governance, and evidence requirements.

Consumer resistance to AI is a design variable, not a constant.

Customer-facing AI draws mild resistance on average, but the response changes sharply with the role, stakes, autonomy, and language surrounding it. Treat acceptance as a portfolio design and sequencing decision, not a blanket reason to accelerate or stop AI deployment.

Data chart

Robot framing produces far more consumer resistance

-0.83AI robots-0.24Algorithm, assistant or system labels

Embodied robot framing produces roughly four times the resistance of other AI labels.

Action guide

  1. Sequence customer-facing AI by domain resistance.Set a slower rollout pace for transportation, legal/public safety, and social welfare than for operational and lifestyle uses.
  2. Choose the AI's label deliberately.Calling it a robot draws far more resistance than calling it an assistant, system, or algorithm, so treat the label as a senior positioning choice across customer-facing products.
  3. Set portfolio limits on visible AI autonomy.Limit how much decision control customer-facing AI visibly exercises because extensive autonomy draws more resistance than limited autonomy.
  4. Refresh resistance assumptions before committing scale.Consumer resistance has weakened over time but remains negative, so base forecasts on current evidence for your own use case, and treat any single average as a starting point, not a general fact about AI.
  5. Treat stated acceptance and observed behavior as separate evidence.The research cannot yet establish that people act more favorably toward AI than they say they will, so a rollout case that assumes stated resistance overstates real resistance rests on an untested claim. Commission your own behavioral read rather than assuming the gap.
  6. Apply this to controlled comparisons of AI versus humans in North American consumer settings.The evidence is experimental responses, not field adoption or revenue, so confirm any rollout case with real usage data from your own market before extending it further.

Evidence

  • Across the full corpus, 66% of studies showed resistance to AI versus humans; the average response difference was small.
  • Robot framing produced roughly four times the resistance of algorithm, assistant, or system framing.
  • Resistance was strongest in transportation, legal/public safety, and social welfare, and mildest in entertainment/lifestyle and in operations and management.
  • Extensive AI autonomy drew resistance; limited-autonomy AI was not reliably different from a human counterpart.
  • The average AI-versus-human response gap narrowed from -0.42 in 2015-2019 to -0.12 in 2020-2025, but still favored humans.
  • Studies measuring what people did, rather than what they said, produced a smaller resistance estimate, but the difference is within the range of chance and the behavior-based estimate is not itself reliably below zero. Field settings and more realistic presentations did not uniformly reduce resistance.
  • The pooled average hides enormous variation: individual results run from strong resistance through to clear preference for AI, and the range a new study would be expected to fall in spans both.

Key takeaway

Consumer resistance to AI is small on average.

Source

Zehnle, M., Hildebrand, C., & Valenzuela, A. (2025). Not all AI is created equal: A meta-analysis revealing drivers of AI resistance across markets, methods, and time. International Journal of Research in Marketing.

Evidence strength: Strong (440 experimental effect sizes comparing consumer responses to AI versus humans across 72 publications, 2002-2023). Generalizes most confidently to experimental, consumer-facing human-versus-AI comparisons in North American samples; less confidently to field adoption outcomes, non-Western markets, and cases where consumers do not know the source is AI. The operations-and-management domain estimate clears only a lower confidence standard than the other domains.